AI chatbot in 3 lines of code.
Stop burning tokens building chatbots from scratch. We did it for you.
Built for indie hackers and SMB sites that need a working chatbot today, not after a 3-week integration.
▶ Try it live — landing page with 6 demos, install code, and comparison
Live demos · API Reference · Roadmap · SKILL Protocol · 日本語 · 中文
import { ChatWidget } from "chatbotlite/react";
<ChatWidget endpoint="/api/chat" title="Acme Plumbing" theme={{ primary: "#1e3a8a" }} />Or plain HTML (Shopify, WordPress, Webflow, anywhere):
<script src="https://unpkg.com/chatbotlite/dist/embed.global.js"></script>
<script>
chatbotlite.mount({ endpoint: "/api/chat", title: "Acme Plumbing" });
</script>Server (/api/chat):
import { ChatBot } from "chatbotlite/client";
import { knowledgeFromFile } from "chatbotlite/node";
const bot = new ChatBot({
knowledge: knowledgeFromFile("./knowledge.md"),
providers: { keys: { openai: process.env.OPENAI_API_KEY } }
});
export async function POST(req: Request) {
const { message, transcript } = await req.json();
return new Response(await bot.replyStream(message, { history: transcript }), {
headers: { "Content-Type": "text/event-stream" }
});
}Add multiple provider keys. If one errors mid-stream, the next provider takes over and the assistant restarts the reply. True token-level replay is on the 0.8 roadmap.
10 providers: OpenAI, Groq, DeepSeek, Gemini, Mistral, Fireworks, Cerebras, SambaNova, OpenRouter, Moonshot.
| 3 lines | React component or <script> tag. No build step needed. |
| 10 LLM providers + local | Auto-failover chain across 10 hosted providers, or point a custom entry at your own Ollama / LM Studio / vLLM. |
| Markdown knowledge | Write services, hours, pricing in a .md file. No vector DB. Anti-hallucination guards built-in. |
| 13 adapters | Stripe, PayPal, Calendly, Cal.com, Formspree, and 8 more. Paste a URL, done. |
| Tool cards | Bot triggers payment, scheduling, file upload, picker buttons inline in chat. |
| Session persistence | Pluggable ChatStorage interface. localStorage default, wire to your own DB. |
| Streaming | SSE tokens render as the LLM types. Streaming cursor in brand color. |
| Defense in depth | Phrase redlines + optional LLM judges for input/output safety. |
| $0/mo forever | Apache 2.0. No SaaS subscription. No per-conversation pricing. |
URL-only adapters. Customer pastes a URL, we open it. Zero backend, zero API keys.
import { stripeLink, calendlyUrl } from "chatbotlite/adapters";
<ChatWidget
endpoint="/api/chat"
tools={{
requestPayment: stripeLink("https://buy.stripe.com/your-link"),
scheduleCallback: calendlyUrl("https://calendly.com/your-page/30min"),
}}
/>Payment: stripeLink, paypalLink, squareLink, lemonSqueezyLink, gumroadLink
Scheduling: calendlyUrl, calcomUrl, savvycalUrl, acuityUrl, msBookingsUrl, googleCalendarApptUrl
Lead capture: formspreeUrl, tallyUrl
The bot emits [SKILL:...] markers. The widget renders interactive cards.
[SKILL:requestPayment amount=4250 currency="cad" reason="deposit"]
[SKILL:scheduleCallback durationMin=15 timezone="America/Vancouver"]
[SKILL:uploadForReview purpose="T4 slip" accept="image/*,application/pdf"]
[SKILL:pickerMessage prompt="Service type?" options="Inspection,Repair,Emergency"]
See the full SKILL Marker Protocol spec.
providers: {
keys: {
openai: process.env.OPENAI_API_KEY,
groq: process.env.GROQ_API_KEY,
deepseek: process.env.DEEPSEEK_API_KEY,
},
chain: [
{ provider: "openai", model: "gpt-4o-mini" },
{ provider: "groq", model: "llama-3.3-70b-versatile" },
{ provider: "deepseek", model: "deepseek-chat" }
]
}Top-to-bottom = priority. Auto-retry on 429/5xx, then fall to next.
A chatbot doesn't need a frontier model. If you already run a local model (Ollama, LM Studio, vLLM, llama.cpp) it speaks the same OpenAI-compatible API, so point a custom chain entry at it — your data never leaves your machine:
providers: {
keys: { openai: process.env.OPENAI_API_KEY }, // optional cloud fallback
chain: [
{ provider: "custom", baseUrl: "http://localhost:11434/v1", model: "llama3.2", apiKey: "ollama" },
{ provider: "openai", model: "gpt-4o-mini" } // falls over to cloud if local is down
]
}apiKey is optional for local servers that ignore auth. baseUrl can also override a hosted provider (e.g. point openai at an internal gateway).
6 verticals, all running on the same npm package: chatbotlite-demos.vercel.app
| Vertical | Demo |
|---|---|
| E-commerce | Bayside Coffee Co. |
| Service business | Acme Plumbing |
| Hospitality | Bella Italia |
| Healthcare | Smile Care Dental |
| Professional services | MaxTax |
| Wellness | Sunrise Yoga |
Building with Claude Code, Cursor, or Cline? Add the chatbotlite-mcp server and your agent can scaffold a working chatbot from a one-line ask ("add a chat widget to my Next.js site, fall back OpenAI → Groq, take Stripe deposits").
claude mcp add chatbotlite -- npx -y chatbotlite-mcpTools: scaffold_chatbot (paste-ready code), list_providers, list_adapters, skill_markers, about_chatbotlite. No API key needed.
- Full API Reference (35KB, LLM-readable)
- llms.txt summary
- SKILL Marker Protocol
- Roadmap
- Design System
- Strategy
PRs welcome. See CONTRIBUTING.md for what we merge fast and what we usually decline.
For security disclosures, see SECURITY.md — please do not open a public issue.
A star helps it reach the next dev who'd otherwise spend the weekend rebuilding a chat widget from scratch. That's basically the entire OSS deal — costs nothing, helps a lot.
Apache 2.0. Use it for whatever, commercial too.
Built by agents-io. Also from our team: Cross-Code Organizer (328★ cross-harness config dashboard — Claude Code, Codex CLI, MCP servers) · PokeClaw (875★).